Multi-frequency sparse Bayesian learning with noise models

Ed Sullivan’s legacy includes significant contributions to the field of signal processing. Inspired by his Bayesian approach, we present results for a method coined Sparse Bayesian learning (SBL) to estimate source parameters. Previously, SBL has been applied to the matched field processing application [K. L. Gemba, S. Nannuru, and P. Gerstoft, “Robust ocean acoustic localization with sparse Bayesian learning,” IEEE J. Sel. Top. Signal Process. 13(1), 49–60 (2019)]. This multi-source scenario required adaptive and robust processing, and included a non-stationary noise model. The adaptive SBL algorithm models the complex source amplitudes as random quantities, providing a degree of robustness to amplitude and phase errors. Further, its formulation is flexible and can accommodate advanced noise models. We consider the application of different noise models in simulations and experimental data and compare SBL performance to traditional processing.